CZ和FZ方法介绍以及硅片的生产步骤2

67International Scientific Colloquium Modelling for Electromagnetic Processing Hannover, March 24-26, 2003Modeling in Industrial Silicon Wafer Manufacturing - From Crystal Growth to Device Processes Th. Wetzel, J. VirbulisAbstract Silicon wafer manufacturing is one of the key processes that determine the yield and the profitability in semiconductor device production. The present paper gives an overview on vari-ous applications for numerical modeling in the wafer manufacturing process. It starts with ther-mal and convection models for the crystal growth process, both by the Czochralski (CZ) and the Floating Zone (FZ) method. Within this field, in particular modeling of electromagnetic field influence on melt flow has become an an indispensable means for the puller design and the process development for both methods. A further chapter is devoted to model approaches for predicting crystal defects, like grown-in voids, self-interstitial aggregates or oxygen precipitates. The defect modeling connect the crystal growth directly to the device manufacturing processes, as crystal defects may be detrimental or beneficial to microelectronic devices, produced on the silicon wafer. A last chapter points briefly to more recent applications of numerical modeling in auxiliary processes, like wafer heat treatment steps, epitaxial growth or wafer cleaning. Introduction During the last five decades, the technology for growing silicon monocrystals from the melt, and the subsequent manufacture of silicon wafers, have become highly developed. The quality, purity and size of today s silicon crystals and wafers have reached an outstanding level. However, it remains time consuming and expensive to develop new processes that meet the crystal and wafer quality requirements for devices with further decreasing design rules. The current transition to a wafer diameter of 300 mm creates further challenges for the wafer industry. One of the main driving forces for the use of numerical modeling in the wafer industry are the extremely high costs for growth furnace design and growth process development. Thermal simulation of the global heat transfer in crystal growth furnaces contributes substan-tially to keeping the development costs at an acceptable level. In particular with the transition to 300 mm diameter CZ crystal growth, magnetic fields have become an important additional means to control heat and mass transfer in the large melt volumes. Numerical modeling facilitates the suitable design of inductor systems and helps to understand the effect of the magnetic fields on the melt flow. However, the growth conditions do not only affect yield or pull rate, but also the quality of the crystals. Grown-in defects like voids, self-interstitial aggregates or oxygen precipitates can be detrimental or beneficial in subsequent process steps. Therefore, defect engineering, i.e. the control of the formation of such defects, is an important part of the whole wafer production. Numerical modeling helps to predict the defect formation and behavior through crystal growth and for subsequent process steps, up to the device processes for microelectronic components in the chip manufacturer ’ s production line. 68Beside the success of the crystal growth and the defect simulation, there is an increasing number of new applications of numerical modeling in the wafer manufacturing industry, e.g. in process steps like wafer polishing, epitaxial layer growth or wet cleaning of the wafers. 1. Crystal Growth There are two major methods for producing silicon monocrystals from the melt. With the Czochralski (CZ) method, today crystals of up to 300 mm are grown. The Floating Zone (FZ) technology is currently scaled up to grow 200 mm crystals. 1.1 Czochralski Growth Fig. 1 shows a longitudinal cut through a CZ growth furnace. The prediction of the temperature field in the whole furnace and in the crystal requires a global thermal simulation, taking into account the heat transfer by conduction, convection and radiation. Software tools for the conduction and radiation calculation along with the pre-diction of the melt-crystal interface are avai-lable from different institutes [1, 2, 3]. They are based on 2D axisymmetric models, em-ploy view factor approaches for the radiation treatment and are based on the Finite Element Method (FEM) or the Finite Volume Method (FVM). Some of these codes allow the reproduction of transient growth processes [2]. With increasing melt volumes and crystal diameters, the consideration of the melt convection has become more im-portant. The melt flow in large diameter CZ crucibles - today diameters of up to 32loads of up to 450 kg are used - is characte-rized by time dependent, three dimensional processes. The time needed for a fully transient, three-dimensional simulation of melt flow in a crucible of the mentioned size however, still prevents their application for industrial engineering purposes. Therefore, 2D axisymmetric steady state models have been developed, that reproduce the most important features of the 3D flow, but do still provide reasonably short calculation time [4, 5]. One of the most critical parts in any CZ melt flow simulation, both 2D and 3D, is the consideration of turbulence. The 2D simulation tools usually employ k- ε or LowRe k- ε or k- ω models. Such tools require careful experimental verification and modification based on comparisons with experiments [5, 6, 7]. A model proposed in [8] provided good agreement of the simulated melt-crystal interface shapes with measured ones for 200 mm crystals. 3D time dependent simulations with LES turbulence models [9] and simulations close to DNS [10, 11] are used to further develop the understanding of the flow behavior and to improve less time consuming simulation models. One goal of such attempts is the global modeling of oxygen transport in the Fig. 1: Schematic layout of an industrial CZ puller and simulated temperature distribution (left) 69CZ furnace, which is extremely sensitive towards any lack of precision in the turbulence modeling. Another aspect, closely related to reliable melt flow prediction, is the modeling of magnetic field influence. In 300 mm CZ growth, different types of alternating (AC) magnetic fields as well as static (DC) magnetic fields are used. The incorporation of the melt flow models into the global heat transfer simulation tools facilitates direct consideration of the effect of such fields on the temperature distribution in the crystal and all coupled phenomena, like point defect dynamics. A 2D axisymmetric model approach for the AC and DC magnetic field effect on the silicon melt is described in [5]. The model has been verified with experimental results from a CZ model setup with various magnetic fields [7, 12]. 1.2 Floating Zone Growth Fig. 2 shows a longitudinal cut through an FZ growth setup [13]. For the numerical simulation of Floating Zone growth there is a complete model chain necessary, starting with the prediction of the phase boundaries at feed rod, molten zone and crystal, based on RF field influence and heat transfer [14]. A second part of the model chain is covering the prediction of the transient melt flow and heat transfer based on 2D [15] and 3D [16] approaches. Based on the melt flow calculation results, the dopant transport can be simulated, yielding finally the macroscopic and microscopic resistivity distributions in grown wafers. The current development of 200 mm FZ crystal growth processes has been substantially supported by the use of such simulation models. Beyond these applications, the influence of additional low frequency inductors has been modeled [14] and considerations about the stress induced dislocation generation have been started [17]. A recently developed improved phase boundary simulation model is presented in [18]. 3. Defect Modeling There are different types of defects in silicon wafers, which are related to the crystal growth conditions and the thermal history of the crystal and wafer. During the solidification process, intrinsic point defects, namely vacancies and silicon self interstitials, are incorporated into the crystal. These intrinsic point defects form grown-in defects during cooling down from melting to room temperature. Impurity atoms can also be incorporated into the growing crystal, later forming precipitates or other defects. The most common of these impurities is oxygen, which is dissolved from the silica crucible in CZ processes and transported through the melt to the crystallization front. Fig. 2: Schematic layout of a Floating Zone setup with simulated features and FEM mesh 70Three types of defects in CZ crystals will be mentioned here specifically: grown-in octahedral voids [19] formed by the aggregation of vacancies (Fig. 3), oxide precipitates [20] and the OSF ring [19]. Whether vacancy or self-interstitial related defects are found in a CZ crystal or crystal region, depends on the value v/G in that region at the crystallization interface during solidification (v being the pull rate, G the temperature gradient at the solid/liquid interface) [19, 21]. In addition to determining these quantities from the thermal simulation, models are used to directly describe the point defect dynamics, i.e. the diffusive and convective transport of these defects during crystal growth as well as their annihilation by recombination [22]. The combination of such simulation models with the thermal simulation results, in particular the transient temperature field in the crystal, allows to design the growth furnace and the growth process in such a way, that grown crystals show a desired defect species. Furthermore, especially for vacancy rich crystals, that are used e.g. for the production of memory chips, it is necessary to ensure a specific size and density of the grown-in voids. Simulation models have been developed, that describe the formation of such defects from vacancies [19]. The size of the grown in voids depends strongly on the cooling rate in a specific temperature interval during cooling down. Therefore, these models do also need the thermal history of the crystal as well as the results of the point defect models. Fig. 4 shows simulated size distributions of voids in fast and slow cooled CZ crystals. The void formation models are also used to predict the impact of wafer heat treatment steps on size and density of the voids. An important part of defect modeling is devoted to the behavior of oxygen in CZ silicon. As oxide precipitates are used as centers for internal gettering [20], models have been developed to predict the size and in particular the density of such defects in CZ crystals and wafers [23]. Another well known defect phenomenon is the so called OSF ring, appearing on the wafer surface after wet oxidation. The nature and formation of the OSF ring has been studied by many authors (see [19] for further ref.). How-ever, there is ongoing research activity in this field, as there are complex interactions of oxygen precipi-tation with other point defects or impurity atoms in the growing crystal, e.g. Nitrogen, that affect e.g. the width of the OSF ring Fig.3: Octahedral void in an as-grown silicon crystal [19] Fig. 4: Simulated size distributions of voids in crystals with different cooling rates 71[24]. 3. Modeling for Wafer Processing Steps Following crystal growth, there are several process steps in a wafer production line. It starts with cutting of the ingots, goes through slicing, etching, polishing and ends with final cleaning of the wafers. Inbetween there can be several other steps like lapping, thermal treatment or epitaxial layer growth. In particular in the areas of etching, cleaning and epitaxial layer growth there are several simulation activities based on Computational Fluid Dynamics (CFD) codes, extended with chemistry and mechanical models. Many questions in this field can be handled already with standard CFD tools, like flow distribution in cleaning baths. Other examples for such simulations are the prediction of temperature distributions in annealing furnaces or in RF heated epi reactors. These more recent applications will gain increasing attention as they open a similar potential for improved physical understanding of the processes as well as for saving con-siderable amounts of time and money for process and equipment development. As in the crystal growth and defect modeling activities, the close collaboration of industry and research institutes will be extremely important for such new modeling topics, to finally have them at the same scientific level and with the same impact on industrial day to day research and development work. Conclusions Numerical modeling is an integral part of today ’ s industrial silicon crystal growth and wafer manufacturing R&D activities, with a variety of applications. Crystal growth simulation with thermal and convection models, both for the CZ and the FZ method is the most well established application. Within this field, in particular modeling of electromagnetic field influence on melt flow has become an indispensable means for the puller design and the process development for both methods. The defect distribution in a wafer is extremely important for the yield and profitability of microelectronic device production lines. Therefore, various model approaches for predicting crystal growth and wafer treatment related defects, like grown-in voids or oxygen precipitates, have been developed and are successfully used in the industry. New applications of numerical modeling in auxiliary processes, like wafer heat treatment steps, epitaxial growth or wafer cleaning are being developed and successfully used in industrial equipment and process design. References [1] Dornberger, E., Tomzig, E., Seidl, A., Schmitt, S., Leister, H.-J., Schmitt, Ch., M ü ller, G.: Thermal Simulation of the Czochralski silicon growth process by three different models and comparison with experiment. J. Cryst. Growth 181 (1997), 461 pp. [2] Van den Bogaert, N., Dupret, F .: Dynamic global simulation of the Czochralski process I / II. J. Cryst. Growth 171 (1997), pp. 65-93 [3] Kurz, M., Pusztai, A., M ü ller, G.: Development of a new powerful computer code CrysVUN++ especially designed for fast simulation of bulk crystal growth processes. J. Cryst. Growth (198/199 (1999), pp. 101-106 [4] Lipchin, A., Brown, R.A .: Hybrid finite-volume/finite element simulation of heat transfer and turbulence in Czochralski growth of silicon. J. Cryst. Growth 216 (2000), pp. 192-203 [5] Wetzel, Th., Muiznieks, A., M ü hlbauer, A., Gelfgat, Y., Gorbunov, L., Virbulis, J., Tomzig, E., von Ammon, W.: Numerical model of turbulent CZ melt flow in the presence of AC and CUSP magnetic fields and its verification in a laboratory facility. J. Cryst. Growth 230 (2001) pp. 81-91 [6] Lipchin, A., Brown, R.A.: Comparison of three turbulence models for simulation of melt convection in Czochralski crystal growt